Actix Fortunes; Explanation of results

[NateBrady23]

Actix fortunes are performing miles ahead of the rest of the field. Check out these results.

@fafhrd91 I imagine that when we post official Round 18 results, there will be a lot of questions about how this was achieved. The framework was already near the top in previous results, but is now performing ~300% better. We suspect that a major contributing factor is a new release of the rust-postgres driver

We're probably going to bring attention to this in the blog post for Round 18. Would you be able to speak to some of the improvements you made?

[fafhrd91]

sure

• new async rust postgres driver. it is rust implementation of psql protocol
• actix is single threaded, it runs in multiple threads but each thread is independent, so no synchronization is needed
• fortunes template uses simd instructions for html escaping
• actix uses generics extensively. compiler is able to use static dispatch for a lot of function calls
• actix uses object pools for requests and responses
• also it uses high performance hash map, based on google's swisstable

[bhauer]

fortunes template uses simd instructions for html escaping

Nice. I am just guessing, but I suspect this is the first use of SIMD instructions in any test implementation.

[zloster]

fortunes template uses simd instructions for html escaping

Nice. I am just guessing, but I suspect this is the first use of SIMD instructions in any test implementation.

The .Net stuff seems to utilize them a lot: CoreCLR (the runtime) and CoreFX (the standard library)

Edit: Also the Java JVM since the middle of Java 8 life cycle (at least). Some details here: https://stackoverflow.com/a/17142855 and the answers after.

[sebastienros]

I am surprised you are not mentioning the usage of pipelining with Npgsql as this is what had the most impact AFAIK.